Particle Swarm Algorithm for Classification Rules Generation

It is the core problem in building a fuzzy classification system to extract an optimal group of fuzzy classification rules from fuzzy data set. A new kind of algorithm is proposed for fuzzy rules´ generating in this work. The idea behind the algorithm is mainly based on both, concepts of data mining and particle swarm optimization (PSO) algorithm. A fuzzy information entropy based measure which generalizes the one used in crisp domain is introduced to measure fuzzy rule´s interestingness, and it combines with other two measures, accuracy and coverage, to construct the composite objective function which is called fitness function in the algorithm. Finally, the algorithm is utilized to solve the famous Saturday morning problem. The result is compared with that of fuzzy decision tree induction method